Or is the normality testing of age/sex in regards to empathy scores and fiction reading interest scores enough?
Based on my judgement alone with the first question, I do not think so. But I still want to make sure.
DETAILS:
I'm testing for 2 things:
1. If fiction reading interest/empathy is dependent on age/sex.
2. if empathy is dependent on fiction reading interest.
*Age and sex are both not approximately normally distributed in regards to both scores of empathy and fiction reading interest based on what I have found.
I know I have to use a non parametric measure for no. 1.
But what do I use for no. 2?
Do I use parametric for 2?
If you would answer non parametric, should I still find the normality distribution of empathy score in regards to fiction reading interest scores, and vice versa?
Or is * an enough justification to use non parametric?
Replies are appreciated.
You should determine whether all your outcome measures are normally distributed first, so that includes fiction reading interest and empathy. Age and sex may be covarying factors but from what I understood from your design, they are not outcome measures.
Age and sex are binary variables and one should not expect them to follow normal distribution. Hence, there is no need to test for normality.
you should use non-parametric statistics to test your hypothesis which usually do not require normality condition.
To answer your question I need some clarification. It seems to me, that what you truly want to know is if fiction reading interest correlates with empathy, age and gender. This is not an experiment, I presume. Therefore, you need to use a regression analysis. Now, what type of regression technique depends on your outcome variable, Fiction Reading Interest. If that outcome variable is a continuous variable, then you just use an old-fashioned Pearson R regression. If your outcome variable is binary (which I don't think makes any sense, I am presuming you are using a continuous scale to measure that), then you would use a Logistic Regression procedure. In either case, PREDICTORS (age, gender and empathy - the latter should also be a continuous scale) do not need to be normally distributed. And, remember, AGE IS NOT BINARY. In fact, age is a true RATIO (you can have somebody with age 0 or 100 with all continuous values in between). Make sure when you are using your stats software to investigate if there are any interactions among predictors.
http://www.theanalysisfactor.com/the-distribution-of-independent-variables-in-regression-models-2/
Thank you Gloriam for highlighting my mistake that Age is a binary variable. Age is an continuous variable. I am sorry for my honest mistake.
No problem. You got it confused with Gender, which is a binary variable. :-)